TL;DR

Anthropic’s product lead Cat Woo shares that product managers in the AI era must shift from “planning six months ahead” to “shipping weekly,” by setting clear goals, shortening the idea-to-delivery pipeline, and building tight cross-functional collaboration to unlock the full potential of AI-native products.

Introduction: Cat Woo at the Center of AI Product Transformation

Cat Woo is the product lead for Claude Code and collaborative work at Anthropic. She works closely with Boris, the technical lead and product visionary—Boris excels at defining what the full AGI will look like three to six months out, while Cat’s primary job is “path planning from the current position to that vision,” including cross-functional coordination, go-to-market, sales, finance, capacity alignment, and clearing launch blockers. About 80% of their ideas align; for the remaining 20%, each pushes forward the parts they care about most.

The Product Manager Role Is Undergoing a Fundamental Shift

From 6-Month Roadmaps to 1-Week (or Even 1-Day) Releases

After interviewing hundreds of product managers, Cat found many still clinging to traditional methodologies. Before AI, technology change was slow; PMs planned in 6–12 month horizons because writing code was expensive and a lot of effort went into coordinating partner teams. Now, AI has dramatically accelerated engineering development and model capabilities improve extremely fast. Many of Anthropic’s product feature timelines have shrunk from 6 months to 1 month, sometimes even to 1 week or 1 day.

Core shift: PMs should focus less on multi-quarter roadmap alignment and more on how to get products out the door as fast as possible—how to turn an idea into something in users’ hands by the end of the week.

Three Things Successful AI PMs Do

1. Set clear goals: Language models are very general, which can lead to ambiguity. Good PMs articulate: “The key user is professional developers, the core problem is notification fatigue from too many permission prompts, the use case is enabling enterprise developers to achieve zero-permission prompting securely.” This eliminates many potential approaches and lets the team focus.

2. Establish a repeatable release process: Almost all Claude Code features ship as “research previews,” clearly marked as early stage, reducing commitment and allowing the team to get something out in two weeks and collect feedback.

3. Help the team build a cross-functional collaboration framework: A tight process exists among engineering, documentation, product marketing, and developer relations. When an engineer thinks a feature is ready and has been used internally, it gets posted to a standing channel; relevant people jump in immediately, and a market announcement can be done the next day. The PM’s job is to set up this process and reduce all friction in shipping.

The Evolution of the PRD: From Document to Metrics and Principles

Cat’s team doesn’t completely eliminate PRDs, but the approach is much lighter:

• Rigorous metrics check-ins: Weekly metric reviews with the entire team, ensuring everyone deeply understands the business’s key goals, trends, and drivers.
• Team principles list: Includes who the key users are, why they’re key, and what the team is willing to trade off. This empowers everyone on the team to make decisions independently without being blocked by the PM or any stakeholder.

Of course, for particularly vague features or projects that require months of infrastructure work, they still write a one-page PRD outlining goals, delightful use cases, and current failure modes that need fixing.

Why Anthropic’s Release Cadence Is So Fast

Someone online created an Anthropic release calendar—almost every day sees a major feature or product. Some speculate it’s because they have the latest and most powerful model (e.g., the rumored unannounced “Mythos”). Cat clarifies: “We’ve been moving quite fast for several quarters now; it’s not entirely Mythos. Mythos is indeed very powerful, and we do use models internally to increase speed, but most of the growth comes from process and team expectations—we want to remove every obstacle to shipping products and ensure everyone can turn their ideas from just an idea into something released to the world in less than a week (sometimes a day).”

Responses to Recent Events

Claude Code Source Code Leak

About a week ago, the entire source code of Claude Code was leaked due to human error—someone used Claude to write a PR and caused a release package update mistake; it got through two layers of human review before being released. Anthropic immediately investigated, confirmed it was a process failure, and has strengthened safeguards to prevent recurrence. The person involved is still at the company, with the focus now on lessons learned.

Open Claude Subscription Restrictions

Anthropic recently blocked third-party products from using subscriptions through Open Claude. Cat explained: Claude demand is extremely high; infrastructure scaling and token efficiency improvements are ongoing. The usage patterns of third-party products differ from first-party products, so to prioritize the experience of first-party products and the API, a difficult decision had to be made. At the same time, subscription users get a certain amount of credits (roughly $200/month in usage subsidies), and Anthropic is doing its best to provide a seamless transition. Cat emphasized: “Companies need to make money, and compute demand is so high that we can’t give it away for free.”

Anthropic’s Product Team Structure

Anthropic currently has about 30–40 product managers, divided into several teams:

• Research Product Team (led by Diane): Responsible for collecting customer feedback on the model, funneling it to the research team, and guiding model releases.
• Claude Developer Platform Team: Maintains the API on which Claude Code is built, and launches things like hosted agents and other ways to build agents.
• Claude Code Team: Responsible for the core product of Claude Code and collaborative work.
• Enterprise Team (and other teams not detailed).

Conclusion

The core competitive advantage for AI-native products has shifted from “planning” to “execution speed.” The product manager’s role is no longer writing lengthy PRDs and driving multi-quarter roadmaps, but defining clear goals, establishing a minimal release process, and enabling the team to iterate in days or even hours. As Cat puts it: “As the cost of writing code has dropped dramatically, what’s more valuable is deciding what to write.”

Source: YouTube video link (https://www.youtube.com/watch?v=PplmzlgE0kg) (@seclink)